src/libpriv: Add kernel-install-integration

[jmarrero]

This is part of the kernel-install integration work done in:

• https://gitlab.com/fedora/bootc/base-images/-/merge_requests/62
• kernel-install: integration #5135
• baseimage: Add kernel-install integration containers/bootc#1018

[jmarrero]

draft because I am not sure if this is actually works, still need to test.

[cgwalters]

So the simplest thing would be do this as part of make install, then it'd end up in the rpm package (and on every system). I believe this change should be safe, but if we do it this way it means it will be invoked (and should just be a no-op) even on dnf (or other) based systems because we'll not have the layout=ostree.

A bit more conservative change would be for us to inject it at postprocess time (hence it only gets injected into ostree systems).

But I'd lean towards the former, and just sanity check that our no-op is working.

[cgwalters]

It's possible the spec file needs adjustment, but hopefully not

[jmarrero]

This breaks kernel-install somehow... I would have expected that the dropin would not get triggered.

dracut[I]: *** Creating image file '/tmp/.tmpU2x9Mg/initramfs.img' ***
dracut[I]: *** Creating initramfs image file '/tmp/.tmpU2x9Mg/initramfs.img' done ***
+ test -f /usr/lib/modules/6.11.4-301.fc41.x86_64/initramfs.img
Error: Process completed with exit code 1.

I need to dig.

[jmarrero]

depmod: ERROR: could not open directory /lib/modules/6.12.7-200.fc41.x86_64: No such file or directory
depmod: FATAL: could not search modules: No such file or directory

It looks like maybe the problem is that the old kernel dir is sticking around... we might be skipping the cleanup for some reason and in my local test I had the DNF config to keep one kernel. re-testing without that.

[jmarrero]

looks like this is wrong:

[2025-01-10T22:15:10.770Z] error: Installed (but unpackaged) file(s) found:
[2025-01-10T22:15:10.770Z]    /lib/kernel/install.d/05-rpmostree.install

[cgwalters]

It's because it was installed to /lib, not /usr/lib (it's a symlink in the real system, but not in the RPM). IMO we should always prefer writing to /usr/lib so change the Makefile.

[jmarrero]

I am not sure why this resolves the container build failing and I double checked and the file is in the correct place. Will do more manual testing before merging this. Because is weird that now even the container-integration tests has no issue removing the old kernel.